rlgym 0.1.2

A python API that can be used to treat the game Rocket League as an Openai Gym-like environment for Reinforcement Learning projects.

The Rocket League Gym

This is a python API that can be used to treat the game Rocket League as though it were an Openai Gym-style environment for Reinforcement Learning projects. This API must be used with the accompanying Bakkesmod plugin.

Requirements

• A Windows 10 PC
• The Steam version of Rocket League (Epic version might be supported but hasn't been tested)
• Bakkesmod
• The RLGym plugin for Bakkesmod (It's installed automatically by pip)
• Python >= 3.7

Installation

Install the library via pip:

pip3 install rlgym

Then simply run example.py from our repo to ensure everything works.

Usage

To run a premade environment, call rlgym.make with the name of the environment you would like to create. For example, the following code will create an instance of the Duels environment:

import rlgym
env = rlgym.make("Duel")

If you would like to build any environment with self-play enabled, include the keyword "self" (not case sensitive) in the name of the environment, like so:

import rlgym
env = rlgym.make("DuelSelf")

You can take a look at example_self.py to see how to handle observations and actions when dealing with more than one agent.

---

RLGym comes with 3 pre-made environments:

• Duels
• Doubles
• Standard

Each can be instantiated by calling rlgym.make with the name of the environment you would like to create.

RLGym also provides you the ability to create your own environments with a number of potential configurations through the custom_args parameter:

env = rlgym.make("Duel", custom_args={
        ep_len_minutes: float,
        game_speed: int,
        tick_skip: int,
        spawn_opponents: bool,
        random_resets: bool,
        team_size: int,
        terminal_conditions: list(rlgym.utils.terminal_conditions.TerminalCondition),
        reward_fn: rlgym.utils.reward_functions.RewardFunction,
        obs_builder: rlgym.utils.obs_builders.ObsBuilder
    })

For more information on how to build a custom RLGym environment, please visit our Wiki.